Spark-X2.5-4B

Name: Spark-X2.5-4B
Author: iFLYTEK

Reasoning modelTool use

Spark-X2.5-4B

Release date: 2026-09-01Views: 6

Live demoGitHub Hugging Face Compare

Parameters

4.112B

Context length

Multilingual

200+ languages

Reasoning ability

3/5

Spark-X2.5-4B is an AI model published by iFLYTEK, released on 2026-09-01, for Reasoning model, with 4.112B parameters, and 1M context length, with a 133.40 score on Gaokao 2026.

Data sourced primarily from official releases (GitHub, Hugging Face, papers), then benchmark leaderboards, then third-party evaluators. Learn about our data methodology

Spark-X2.5-4B

Model basics

Reasoning traces

Supported

Thinking modes

Thinking Mode (Default)Standard Mode

Context length

1M tokens

Max output length

No data

Model type

Reasoning model

Modality (in / out)

Text → Text

Release date

2026-09-01

Model file size

About 7.7GB (BF16 safetensors)

MoE architecture

Total params / Active params

4.112B / Not applicable

Knowledge cutoff

No data

Spark-X2.5-4B

Open source & experience

Code license

Apache 2.0

Weights license

Apache 2.0- Commercial use permitted

GitHub repo

github.com/XHToken/Spark-X2.5

Hugging Face

huggingface.co/XHToken/Spark-X2.5-4B

Live demo

N/A

Spark-X2.5-4B

Official resources

Paper

N/A

DataLearnerAI blog

N/A

Spark-X2.5-4B

API details

API speed

5/5

No public API pricing yet.

Spark-X2.5-4B

Benchmark Results

Spark-X2.5-4B currently shows benchmark results led by IF Bench (10 / 35, score 75), GPQA (6 / 17, score 67.40), τ²-Bench (22 / 44, score 75.10). This page also consolidates core specs, context limits, and API pricing so you can evaluate the model from benchmark results and deployment constraints together.

General Knowledge

2 evaluations

Benchmark / mode

Score

Rank/total

GPQA

Thinking Mode

67.40

6 / 17

HLE

Thinking Mode

12.30

157 / 189

Coding and Software Engineer

4 evaluations

Benchmark / mode

Score

Rank/total

SWE-bench Multilingual

Thinking ModeTools

53.30

28 / 28

SWE-Bench Pro - Public

Thinking ModeTools

44.40

51 / 60

SWE-bench Verified

Thinking ModeTools

41.60

103 / 115

SciCode

Thinking Mode

34.70

15 / 15

Agent Level Benchmark

4 evaluations

Benchmark / mode

Score

Rank/total

τ²-Bench

Thinking ModeTools

75.10

22 / 44

Workspace-Bench

Thinking ModeTools

31.20

1 / 1

τ³-Bench

Thinking ModeTools

30.40

1 / 1

VitaBench 2.0

Thinking ModeTools

25.20

1 / 1

Instruction Following

2 evaluations

Benchmark / mode

Score

Rank/total

IFEval

Thinking Mode

1 / 1

IF Bench

Thinking Mode

10 / 35

AI Agent - Information Search

1 evaluations

Benchmark / mode

Score

Rank/total

BrowseComp

Thinking ModeToolsInternet

40.90

49 / 55

Math and Reasoning

4 evaluations

Benchmark / mode

Score

Rank/total

Gaokao 2026

Thinking Mode

133.40

1 / 1

AIME 2026

Thinking Mode

90.70

15 / 20

HMMT Feb 2026

Thinking Mode

81.20

1 / 1

IMO-AnswerBench

Thinking Mode

74.20

22 / 24

Long Context

1 evaluations

Benchmark / mode

Score

Rank/total

AA-LCR

Thinking Mode

56.30

27 / 28

AI Agent - Tool Usage

3 evaluations

Benchmark / mode

Score

Rank/total

BFCL-V4

Thinking ModeTools

65.10

1 / 1

MCP-Atlas

Thinking ModeTools

54.60

38 / 41

MCPMark

Thinking ModeTools

14.20

1 / 1

View benchmark analysis Compare with other models

Spark-X2.5-4B

Publisher

iFLYTEK

View publisher details

Spark-X2.5-4B

Model Overview

Spark-X2.5-4B is an AI model published by iFLYTEK, released on 2026-09-01, for Reasoning model, with 4.112B parameters, and 1M context length, with a 133.40 score on Gaokao 2026.

DataLearner on WeChat

Follow DataLearner on WeChat for AI model updates and research notes.